The Korea Local Administration Review

Local governments are confronted with rapidly changing environmental dynamics, including demographic shifts due to low birth rates and an aging population, the emergence of COVID-19, and global economic stagnation. The accuracy of tax revenue forecasting is gaining attention for stable financial management while reflecting the environmental changes surrounding local governments. In this study, we collected data on local tax collections from 1994 to 2021 and applied various algorithms to predict total local tax revenue from 2017 to 2023. The results were compared with actual collections and existing estimation methods to determine if machine learning methodologies can be applied to local tax revenue forecasting, and if so, which models (algorithms) have the highest accuracy. Accuracy was evaluated using MAPE and RMSE. We found that neural network-based machine learning methodologies are more accurate. This confirms the usefulness of applying machine learning methodologies to improve the accuracy and speed of local tax revenue forecasting. The significance of this study is that it explored improvement measures to reduce the error of local tax revenue estimation by applying various machine learning methodologies based on data.